[SOLVED] Proxmox Shell for Upgrades logs in wrong directory

[Urbaman]

Hi,

I'm facing a strange behaviour.
I have a 3xNodes cluster working find for month (almost a year and a half), now it begins to have this behiviour:

On Node1, when I start the shell from the Updates to Upgrade, it logs in / directory instead of /root
On Node2 and Node3 it rightly logs in /root

On all of the three nodes, if I start a Shell, it rightly logs in /root

On all of the three nodes, if I SSH, i trightly logs in /root

So, why only for the Updates-Upgrade shell on just one node it fails to log in the right directory?

Thank you.

 

[Urbaman]

Hi,

Any one having this problem?

Why on earth the two shells on the same node (console and upgrade shell) put me in a different path?

 

[Urbaman]

For reference, the root home directory seems to be right, here are three ways to access the console:

ssh (right):



Gui Shell (right):



Gui Shell from upgrades (wrong):



As said, the other nodes are really ok.

Thank you.

 

[Urbaman]

Hi, any one know where I can see or change the directory where the Upgreades shell starts?

 

[RolandK]

i think this is as side effect on how console/shell is spawned internally. i don't think this can be changed.

why is this a problem for you ?

to explain what you see - the parent process (in both cases) has it's homedir set to "/" and sub-processes are inheriting that, as long they don't change the dir (what login -f root does)

# pstree -a -p -g|grep -A2 termproxy
  |   |       `-termproxy,3249282,3249281 5900 --path /nodes/pve-pc1 --perm Sys.Console -- /usr/bin/pveupgrade --shell
  |   |           `-pveupgrade,3249285,3249285 /usr/bin/pveupgrade --shell
  |   |               `-bash,3249369,3249369 -l
--
  |           `-termproxy,3255725,3255724 5900 --path /nodes/pve-pc1 --perm Sys.Console -- /bin/login -f root
  |               `-login,3255727,3255727 -f
  |                   `-bash,3255732,3255732


# ls -la /proc/3249282/cwd
lrwxrwxrwx 1 root root 0 Feb 11 14:26 /proc/3249282/cwd -> /
# ls -la /proc/3249285/cwd
lrwxrwxrwx 1 root root 0 Feb 11 14:35 /proc/3249285/cwd -> /
# ls -la /proc/3249369/cwd
lrwxrwxrwx 1 root root 0 Feb 11 14:35 /proc/3249369/cwd -> /
# ls -la /proc/3255725/cwd
lrwxrwxrwx 1 root root 0 Feb 11 14:26 /proc/3255725/cwd -> /
# ls -la /proc/3255727/cwd
lrwxrwxrwx 1 root root 0 Feb 11 14:36 /proc/3255727/cwd -> /
# ls -la /proc/3255732/cwd
lrwxrwxrwx 1 root root 0 Feb 11 14:36 /proc/3255732/cwd -> /root

 

[Urbaman]

It actually does what it must do: run upgrades, so no actual problem.

I'd like to understand why it is so, and it is so on just one node, to prevent/solve moments in which it could be a problem (say it begins doing it on normal GUI shell, and not reading env settings from the right dir).

 

[RolandK]

that it's different on one node is weird indeed. please double check if you're not observing wrong

i showed you how to investigate, i.e. how to determine what's the current working directory of a process, so you can analyze this

 

[Urbaman]

Ok,

Thank you for the insight, I found out it works different if I update a node I'm directly connected to on GUI (node1) or not (node2):

pstree -a -p -g|grep -A2 termproxy
  |       |   `-termproxy,2386260,2386259 5900 --path /nodes/pvenode1 --perm Sys.Console -- /usr/bin/pveupgrade --shell
  |       |       `-pveupgrade,2386262,2386262 /usr/bin/pveupgrade --shell
  |       |           `-bash,2386666,2386666 -l
--
  |           `-termproxy,2388437,2388436 5900 --path /nodes/pvenode2 --perm Sys.Console -- /usr/bin/ssh -e none -t 10.0.100.12 ...
  |               `-ssh,2388439,2388439 -e none -t 10.0.100.12 -- /usr/bin/pveupgrade --shell
  |-pvefw-logger,1727664,1727664
--
  |           |-grep,2389326,2389325 -A2 termproxy
  |           `-pstree,2389325,2389325 -a -p -g
  |-systemd,2387104,2387104 --user

Everytime I upgrade from gui, it just launches pveupgrade directly if I'm on the right node, but it goes ssh on other nodes, so that's why they run from different directories.

I think I'm fine with it, knowing it's substancially a pveupgrade thing between the gui-connected node and the other cluster nodes, even if a more coherent behiviour could be better cluster and feature-wide.

Thank you very much again!